OKUN S LAW IN MALAYSIA: AN AUTOREGRESSIVE DISTRIBUTED LAG (ARDL) APPROACH WITH HODRICK-PRESCOTT (HP) FILTER

Transcription

1 OKUN S LAW IN MALAYSIA: AN AUTOREGRESSIVE DISTRIBUTED LAG (ARDL) APPROACH WITH HODRICK-PRESCOTT (HP) FILTER Ngoo Yee Ting i and Loi Siew Ling ii Okun s Law is common and existed in most of the European countries. It explained the negative relationship between output growth and unemployment rate. It has been a norm of the existence of the negative relationship between output growth and unemployment. The impact of high unemployment rate is direct, right after the recession or the decrease of output growth. The importance of finding the effect of the relationship will determine the effectiveness of the correct policy. This paper tends to examine the existence of Okun s relationship in Malaysia context. The relationship is measured by applying the first difference and gap model with Hodrick-Prescott filter (HP filter), furthered with Autogressive Distributed Lag (ARDL) to determine the co -integration between the variables and their causality. Field of Research: Growth & Innovation Keywords: Okun s Law, Difference-gap model, HP filter, ARDL, causal relationship. i Ngoo Yee Ting, Faculty of Accountancy & Management, Lecturer, UTAR. ngooyt@utar.edu.my ii Loi Siew Ling, Faculty of Economy and Management, PHD candidate of University Putra Malaysia. loisiewling@hotmail.com 1

2 1. Introduction Ever since the financial crisis attack in 2008, Malaysia s economic growth faces negative growth, of which started at the beginning of third quarter of 2008 and sustained until the second quarter of 2009 where there seems to be some dim lights in the gloomy days. The financial crisis has caused negative output to the country s growth, hence worsened the unemployment situation where the unemployment rate increased tremendously. This situation also happens vice versa where the increase of unemployment rate will also affect the output growth in a country. Therefore, the Okun s law is the best tool to examine and explain the negative relationship between output growth and unemployment. The initial Okun s law explained the negative relationship between output growth and change in unemployment rate. A negative output growth will cause a high rise in unemployment and vice versa. Over the time, Okun s law has revolved from the difference version to gap and dynamic version. The coefficient found from the estimation will enable one country to determine the effect of output growth to unemployment. The lower the coefficient, the higher security of the job will be when the recession occurs. Alternatively, one may be able to determine the effect of unemployment to output growth depending on one s objective of findings. The main concern of the recession is the negative growth effect towards unemployment in the country. Without job, without the income, people find life difficult to survive. Tentatively, we would assume the unemployment will increase when there is negative growth in the country. It is therefore important to analyse if the effect of the changes in the country s growth to the changes of unemployment is damaging to the citizen of Malaysia in which the appropriate policies were needed. 2

3 Malaysia has undergone the typical negative relationship of output growth to unemployment. Malaysia faces positive effect of unemployment when the economic transformed from agriculture sector to manufacturing sector, negative effect when there is a world wide recession or crisis and positive again when the economy recovered from the phenomena. The focus of this paper is to examine the existence of Okun s relationship in Malaysia context. The relationship is measured by applying the first difference and gap model with HP filter (Hodrick-Prescott filter), furthered with ARDL (Autogressive distributed lag) to determine the cointegration between the variables and their causal relationship and the stability of Okun s Law in the long run. 2. Literature review Knotek (2007) noticed three ways of finding the Okun s Law. First, the difference version of Okun s Law captured the changes in the unemployment rate from one quarter to the next quarter. Second, the gap version captures unemployment to the gap between potential output and actual output, in which was also applied by Thirlwall (1969) in calculating natural rate of growth and the third, the dynamic version implied the past output, current of output and past unemployment to affect the unemployment. Last but not least, the production-function version which combined the theory of production function (labour, capital and technology) with the gap version of Okun s Law. Amornthum (2002) classified the finding of potential output to statistic approach and economic approach. Statistical approach will find potential output by extracting the cyclical trend from the business cycle applying Hodrick-Prescott (HP) filter. HP filter is commonly used to determine the business cycle dynamics which is also applied by Lee (2000) in the research. Economic approach focus more on the economic theories and the most common method will be the production function approach where Okun was the first to estimate the potential GNP. Petkov (2008) enhanced the method combining both statistical and economic approach by applying HP into Autoregressive Distributed Lag 3

4 Approach (ARDL). HP was applied in order to capture the Non Accerlerating Inflation Rate of Unemployment (NAIRU). This was furthered with error correction model (ECM) and the final finding was that although the Okun s coefficient may changes over time, the model continues to show strong relationship between output growth and unemployment. In other word, this model falls into the dynamic version of measuring output suggested by Knotek. However, Knotek explained the dynamic relationships using the rolling regression method, applying a sample of 13 years data for each rolling regression. NAIRU measurement was also applied by Vougas (2003) to determine the natural rate unemployment at Greece to estimate Greek unemployment using dynamic model as well. However, the NAIRU was estimated from augmented Phillips curve with past inflation adding to with the hysteresis theory, the previous unemployment rate to determine the natural rate of unemployment, revealing a similarity of dynamics version. Although Amornthum found that the Okun s coefficient may changes over time, Lee (2000) discovered that the coefficient changes over different countries. Lee supports the validity of Okun s law, but not the robustness of Okun s law application of other OECD countries. The research applied the first difference version and gap version, followed by HP filter, the Beveridge-Nelson decomposition procedure, and the Kalman filter based on the NAIRU framework. Thus, the research found strong evidence of structural change in the Okun relationship. However, when Lee expanded the model to the error-correction framework and allowing the asymmetric effects, the relationship of the Okun s Law is not as robust as reported by Okun in his finding. Additional to that, the estimates were different from one country to another between US and OECD. Where the innovation and cyclical were concerned, Evans (1989) and Weber (1995) used a bivariate VAR model to explain the dynamics between output and unemployment. The model was used to estimate the innovations and the cycle trend. Besides adopting bivariate VAR, Weber (1995) also compared the result with the other three different methods of estimation (static Ordinary Least Squared (OLS), cointegrating regression and ARDL). Weber found that the estimates are sensitive to which method is adopted. It 4

5 is found that the coefficient is rather sensitive to the method used to estimate cyclical output and unemployment. The static estimates supported Okun s original estimate but the dynamic estimation suggested smaller value for the coefficient. Okun s (1962) findings pave an important background on the analysis of the relationship between output growth and unemployment. The initial finding of Okun s law predicts that if output were to increase by 3 percentage point, it will reduce the unemployment rate by 1 percent. This was agreed by Weber (1995) but he found smaller value of coefficient if dynamic specifications were applied. However, when applied to Malaysia, Mohd Noor, Zaleha., Mohamed Nor, Norashidah and Abdul Ghani, Judhiana (2 007) found the existence of Okun s law in Malaysia with coefficient of -1.75, significant at 1 percent level which is lesser than the original Okun s law and the model is not stable Additional to that, they also found two-way causality between unemployment and output growth in Malaysia. Thirlwall (1969) found that neglecting the estimation of natural rate of growth over a past period or not, will show equal result. The natural rate of growth can de derived directly from the same regression. Thirwall commented that the labour hoarding in Okun s law can be solved by reversing the dependent and independent variables in the regression equation. However, this comment was critiqued by Monhollon and Cullison (1970). They argued that the least squares regression technique will have a biased estimation. Additional to that, Okun s Law also neglected the re-entry of discourage worker effect in the labour hoarding element but Thirlwall s natural rate of growth technique was considered consistent as the result was applicable for long run where the short run effect can be avoided and this included the short term effect of labour hording and re-entry of discouraged workers. Although Okun s Law applies to most of the country, but according to Prachowny (1993) Okun s law is insufficient in the measurement of the efficiency of production. Prachowny found that labour input as well as other factors of production especially the changes in weekly hours and capacity utilization will have independent effects on the output gap. 5

6 In conclusion, Okun s Law has transformed from the simply equation between unemployment and output growth into a dynamic model and was applicable for the forecast of the relationship in policy making. Although there might be limitations of using Okun s law, but it provides a basic background of the relationship between the two variables. 3. Theoretical Background Arthur Okun believes the adverse relationship between output and the unemployment rate is not a simple relationship; Okun s law attempted to relate the change in the unemployment rate to the deviation of the rate of output growth from the normal growth rate. The Okun s law specification estimated is: u u ( g g ) (Eq 1) t t 1 yt y where u t is the unemployment rate in year t, u t-1 the unemployment rate in year t-1, g yt the growth rate of output from year t-1 to year t and g y is the normal growth rate. According to Okun, the unemployment rate decreases about 1 percentage point for every 3 increases in real GDP. However the later research and data have shown that the relationship between output and unemployment is not as stable as what Okun s law has predicted. 4. Description of Variables and Data Analysis ln( u t ) ( GDPgap) (Eq 2) The data set used for this study is sourced from Bank Negara Malaysia s Malaysia quarterly data from 2000 (I) to 2007(IV). The u t is the unemployment growth rate. The national income gap is the output gap which is estimated using Hodrick-Prescott filter. This approach is based on a linear filter aimed at removing low frequency variation from a series. HP filter is also a weight parameter of which rules the smoothness of the resulting trend line. All variables are entered in natural log form and the choice of log formulation permits direct estimation of the elasticity. 6

7 5. Methodology and Analysis The objective of this article is to employ cointegration and error-correction modelling to test the causal relationship between unemployment rate and real national income growth rate by using quarterly data for Malaysia from year 2000 to We use a relatively new estimation technique, which is the bounds testing approach to cointegration, within an autoregressive distributive lag (ARDL) framework, proposed by Pesaran and others (Pesaran and Pesaran,1997; Pesaran and Shin,1999; Pesaran et al., 2001). The main purpose to employ ARDL in this study is mainly due to the order of integration is no longer a sensitive issue, it can be applied regardless of whether the regressors are I(0) or I(1), as normally the nature of national income is non stationary. In addition, the bound test stressed good small sample properties in comparison to standard cointegration analysis and without the pre conditions for stationarity. The following autoregressive distributed lag, ARDL [p,q] model will be estimated in order to test the cointegration relationship between unemployment rate and national income. A three-stage procedure was followed to test the direction of causality. The starting point for testing the long run relationship among variables is preceded by tests for the integrated order of series as only variables integrated in I(0) and I(1) may be cointegrated. Augmented Dickey Fuller (ADF) and KPSS unit roots were employed where the null hypothesis of ADF is a series is I(1) while null hypothesis of KPSS is a series is I(0). If the order of integration variables is zero or one, we can confidently apply ARLD to test the cointegration between unemployment and national income gap. In this stage, it involves testing for the existence of a long run equilibrium relationship between unemployment rate and national income gap within a multivariate framework. This involves examining the existence of a long-run relationship using the following unrestricted error-correction model (UECM): UN t p q UN t Y t Yt i UN t i Eq t i 1 i 1 7

8 ARDL approach consists first in specifying and estimating a general distributed lag model. In order to justify retention of lagged level variables in Eq 3 which implies cointegration among them, Pesaran et al. (20 01) proposed applying the familiar F-test with new critical values that they tabulate, followed by a joint significance test, where the null and alternative hypotheses are as below: H O : γ 1 = γ 2 =0, H A : γ 1 γ 2 0. Two asymptotic critical value bounds provide a test for cointegration when the independent variables are I(d) with 0 d 1. The lower bound assumes that all the repressors are I(0), and the upper bound assumes that they are I (1).If the computed F- statistics is less than lower bound critical value, it does not reject null hypothesis of no cointegration. However, if the computed F-statistics is greater than upper bound critical value, then it rejects null hypothesis, which means that there are steady state equilibrium between the variables in the model. This leads to pinpoint potential structural breaks and establish the suitable significant lags in the variables. First unit estimation shows that the growth rate of unemployment rate is stationary data, however the national income gap variables has unit root problems at I(0) and is stationary at first difference level. Table 1 Result of Unit Root Test Level Form Constant ADF Constant with trend Constant KPSS Constant with trend UN *** *** Y * ** First Different Constant ADF Constant with trend Constant KPSS Constant with trend UN *** *** Y *** *** Notes: The asterisk indicates the following levels significance: *** 1%, **5% & * 10%. 8

9 From the Table 2, the results of ARDL bound test are shown. According the computed F statistics, null hypothesis of no cointegration in lag order 1, 2 and 3 are rejected at 1 percent significance. The computed F-statistics for lag order are above the upper and bound critical value. Hence, the presence of a long run cointegration is occurring in between the UN and its determinants are confirmed based on the result of bounds testing. As suggested by AIC and Durbin Watson stat, lag order 1 was chosen. Table 2: The Result of the F-test for Cointegration Lag Order on Each First Differenced Variable UN *** *** *** AIC Durbin Watson Stat Notes: The critical value of F-statistics for lower bound and upper bound are and respectively, at the 5% significance level. The critical value of F-statistics for lower bound and upper bound are and respectively, at the 1% significance level Sources from Pesaran et al. (2001, p. 300), Table CI(iii) Case III unrestricted intercept and no trend. The asterisk indicates the following levels significance: * 5%, **1% Table 3: Result of Long Run Elasticities of Okun s Law in Malaysia Normalized Variable C Y U *** The asterisk indicates the following levels significance: * 5%, **1% The Okun coefficient in Table 3 also shown the immediate impact of changes in the national income gap is negative and significant at the 1 percent level which means that 1 percent decrease in output growth will increase unemployment rate by percent. Next, we construct Granger causality test augmented with a lagged error correction term where the series are cointegated. Given the ARDL test suggest that UN and Y are cointegrated. We augment the Granger-type causality test with a lagged error correction term. Thus, the Granger causality test involves specifying a multivariate VECM in lag order one. 9

10 The existence of a cointegrating relationship among UN and Y suggests that there must be Granger causality in at least one direction, but it does not indicate the direction of causality between the variables. Table 4 examines short run and long run Granger causality within the Error-Correction Mechanism (ECM). The F statistics on the explanatory variables in each of the three equations indicates the statistical significance of the short-run causal effects. In the equation when Y is the dependent variable, the t- statistic on the coefficient of the lagged error-correction term indicates the statistical significance of the long-run causal effect. Table 4: Result of Granger Causality Dependent Variable UN Y ECT UN *** *** ( ) Y *** *** ( ) Notes: a The Wald statistic, which test the join significance of the lagged values of the independent variables is reported. This statistic is to be compared with F-statistics. b the t-statistics reported in parentheses. The asterisks indicate the following levels of significance: * 10%, ** 5% and *** 1 % The causality relationships between UN and Y are summarised in Table 4. From the Wald test statistics that exceed the critical F values, implying the null hypotheses of the independent variables do not cause the dependent variable. The ECTs in Table 4 indicated that there exists a mechanism in correcting the disequilibrium between unemployment rate and national income. This finding, therefore, reconfirms the long run stable Okun Law relationship. We also can confirm the bi-directional causality in Okun s equations and the test for cointegration. 6. Concluding Remarks We empirically supported that for quarterly data and with the use of ARDL approach, Okun s law applies in Malaysia, and the Okun s coefficient is Likewise, VECM causality tests in the Granger sense run in a bidirectional way between unemployment and national output. 10

11 The high coefficient of may be due to the HP filter which rules the smoothness of the resulting trend line and thus ignore the natural rate of unemployment and natural rate of output. Additional to that, the job securities in Malaysia is low and the tendency of recession to job loss is high. There is also a trend of jobs offered in contract based rather than permanent. The need for the correct policies to counter recession is important knowing the effect on unemployment is rather severe. The methods used and the results presented in this paper provide insights into the effects on the unemployment rate of deviations of national output. This evidence supports the results of other authors for different series and periods, which allows us to use it as a good instrument of analysis and forecasting of the economic cycle; allows us to estimate the sacrifice rate of long run unemployment; and moreover, shows that the economic policy must focus by all means in avoiding fluctuations in growth and at the same time reducing unemployment, because this way it will stimulate growth in the long run. 11

12 References Amornthum, Somchai. (2002). Japan s Potential Growth: An HP Filter Approach. (Research Paper for Econ614, Department of Economics, University of Hawai I at Manoa). Edward S. Knotek, II. (2007). How useful is Okun's law?. Economic Review. Federal Reserve Bank of Kansas City, Fourth Quarter, Mohd Noor, Zaleha., Mohamed Nor, Norashidah and Abdul Ghani, Judhiana. (2007). The Relationship Between Output And Unemployment In Malaysia: Does Okun's Law exist?. International Journal of Economics and Management, 1(3), Lee, Jim. (2000). The Robustness of Okun s Law: Evidence from OECD Countries. Journal of Macroeconomics, 22, Monhollon, Jimmie R. and Cullison, William E. (1970). Okun s Law and the Natural Rate of Growth: Comment, Southern Economic Journal, 37(2), Okun, A.M. (1962). Potential GNP: Its Measurement and Significance, Proceedings of the Business and Economic Statistics, Pesaran, M.H., Shin, Yong Cheol and Smith, R. (2001). Bound Testing Approaches to the Analysis of Level Relationships. Journal of Applied Econometrics, 16, Petkov, Boris. (2008). The Labour Market and Output in the UK Does Okun s Law Still Stand? Discussion Papers Bulgarian National Bank, DP/69/2008. Prachowny, M.F.J. (1993). Okun s Law: Theorectical Foundations and Revisited Estimates, Review of Economics and Statistics, 75(2), Schreft, Stacey L. and Singh, Aarti. (2003). A Closer Look at Jobless Recoveries. Economic Review. Federal Reserve Bank of Kansas City, Second Quarter, Thirlwall, A.P. (1969). Okun s Law and the Natural Rate of Growth, Economics Journal, 36(1), Southern Vougas, Dimitrios V. (2003). Unemployment in Greece. Journal of Policy Modelling, 25, Weber, Christian E. (1995). Cyclical Output, Cyclical Unemployment, and Okun s Coefficient: A New Approach, Journal of Applied Econometrics, 10,